Abstract
The early decades of sound recording are a period of rapid change in both recording
technology and performance practice. Primitive recording equipment, minimal
documentation and musical pitch varying by time and location make it difficult for
archivists to know the proper playback stylus size and playback speed. Using four
very large data sets of both pristine and worn media, combined with a highly
consistent selection methodology, this study investigates the level of accuracy
attainable when selecting stylus size and determining playback speed. The analysis
shows it is possible to determine, with high certainty, the correct playback stylus
size per record label during this period of early sound recording. The analysis then
shows that the distribution of pitch follows a Gaussian distribution of randomness;
meaning, it is not possible to know with certainty the proper playback speed.
Commentary discusses the cultural environment of the early recording era and the
implications for archival practice today.
Historical Context
Every field of critical inquiry that engages with the past inevitably must address
questions that are very difficult, if not impossible to answer. What color were
dinosaurs? What was before the Big Bang? On what day was Henry V born? Was
Thomas Jefferson a good violinist or a bad violinist? Aristotle wrote, "Everyone
desires to know”. Our desire for knowledge and certainty propels human inquiry,
especially when that inquiry involves the roots of our identity. The insatiable need
to know leads us to uncover the previously unknown, even though the answer, if
ultimately found, may prove dissatisfying, either because it challenges our
assumptions, or is simply not very interesting after all. Furthermore, we must
embrace and apply new information and knowledge appropriately; and accept when
we do not and may not ever know certain information, and accept that our actions
are based on assumptions and opinion, rather than facts, however well founded.
On the front of the Stephansdom Cathedral in Vienna are 2 iron bars, each about a
meter long, one about a hands’ width longer than the other. Installed in the Middle
Ages, these were the official Viennese measurements of a yard for use by local and
traveling merchants for general commerce in the city. There was a measure for
linen, and a separate measure for drapery. A traveling merchant adjusted his
practice to the local standards. For hundreds of years, this was the state of
standardization in the world.
The early part of the 20 th century was a time of great changes. The Industrial
Revolution brought prosperity to a new middle class. New modes of transportation
brought far-flung parts of the world into contact. As manufacturing expanded with
the introduction of interchangeable parts, the need for standards in weights and
measures became apparent. Likewise, manufacturers soon found ways to create
entertainment goods for the new mass market.
At the same time, many technologies were new and poorly understood. Electricity
was lighting homes before anyone understood grounding. Electric motors and
microphones 1 were crude. Their use in sound recordings was limited prior to the
development of electrical recordings ca. 1923. Speed was hard to measure and
regulate 2 . There was no understanding of, much less the means to measure,
distortion. There were no definitions for frequency response and level, speed
fluctuation, gain, or linearity.
Standards as we understand and use them today, were taking root in the early
decades of the 20 th century. As the formerly isolated peoples of the world began to
interact more frequently, they shared their goods and cultures. To work together
they needed to adapt or adopt the things they wished to share. For musicians to play
together they needed a common pitch and tuning system. Between 1895 and 1910
there were three international meetings to define the frequency of the A above
Middle C. At those meetings the standard rose from 430 to 435, then to 440Hz 3 . It
takes a long time for a standard like this to become widely practiced. Musicians, like
people in all walks of life, can be slow to adopt change. More practically, instruments
are not replaced very frequently. Church organs and string instruments may last
centuries 4 . The physical dimensions of an instrument affect its characteristic pitch.
Into this brave new world of evolving standards, fluid pitch, and new inventions
arrives recorded sound. One might reasonable say the early technicians and
recording engineers were "making it up as they went along”. For some
experimenters we have lab notes that document the goals, methods, and discoveries
of their work 5 . For some recording sessions, the recording staff kept notes. Thomas
1 The microphone was developed independently by Emile Berliner and David Edward
Hughes in the 1870s. It is one of the fundamental precursors of Alexander Graham Bells’
2 It’s one thing to make an electric motor go 'round. It another thing to make it go 'round at
a consistent speed, an imperative for sound recordings.
3 The issue of pitch would remain important. When the new International Standards
Organization (ISO) was formed in 1955, one of their first tasks was to adopt a uniform
frequency of A=440 as ISO 16. This is where we get the expression "Standard Pitch”, by
which we mean A=440.
4 Woodwind and brass instruments rarely last more than a generation. Professional
musicians may "blow out” an instrument in 5-10 years. This idea was put to me by Jay
Krush, founding member of Chestnut Brass Company, an ensemble that specializes in
playing early brass instruments. Most of the historic instruments they've been able to
collect are real dogs. "The good ones were played until they died. The ones that survive
were owned by amateurs who rarely played them, then threw them in a closet until their
grandchildren found them."
5 An excellent summary of work by the Edison staff: STYLUS SHAPES and SIZES: Preliminary
Comments on Historical Edison Cylinder Styli, by Bill Klinger, Chair of the ARSC Cylinder
Committee.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
2
Edison left behind some very detailed technical drawings. Unfortunately, just as
they were "making it up as they went along”, they didn’t really know what
information might be worthwhile to keep - either for the benefit of their developing
profession, or for future researchers.
Despite the many limitations of weight driven lathes and large acoustic horns, and
the uncertainties about practices in the early decades, we nonetheless have
documents of the cultural record that do not exist from earlier times. Audiovisual
records are what distinguish the cultural record of the 20 th century. This is what
drives our curiosity to understand these media. We gain both a better
understanding of our history during this important period of change, and as
audiovisual archivists, an appreciation for the roots of our profession.
The Problem
There are serious challenges to determining the correct way to reproduce these
recordings. At what speed was the lathe turning on the day this particular recording
was made? What pitch did the musicians tune to? How accurately regulated was the
speed of the lathe? Did the musicians re-tune between takes? Were the musicians
from different traditions and choose a compromise pitch for the recording date? 6
As we can see, there are many contextual problems to determining "what is the
correct speed” of an acoustic 78rpm disc. Lacking strong documentation from a
period of general uncertainty, we are forced to work with the recording engineer’s
final product, that is the discs themselves. The overwhelming majority of the discs in
existence are heavily worn from many years of use and enjoyment. This wear means
our work in reproduction, especially stylus selection, must take this deterioration
into account in the data we gather. Just as archaeologists must interpret fossils and
astronomers must compensate for the affect of the Earth’s atmosphere when
peering into the heavens, record wear obscures our data 7 .
6 The author is grateful to Clara Blood, DMA, for furnishing this fascinating discussion on
this topic as viewed from a different vantage: Oboists as keeper of orchestral pitch.
http://www.oboeclassics.com/Burgess.htm
7 This assumption regarding the impact of record wear is brought into question later in this
paper.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
3
The Digitization Process
The playback and digitization process followed the widely adopted best practices of
the trade:
• Hand-delivered from home institutions
• Cleaned with Keith Monks Record Cleaning Machine (distilled water)
• Pictures of the label and matrix (Nikon D810 at 400ppi TIFF)
• Technics SP-15 Turntable fitted with needles by Expert Stylus
• 4 KAB preamps
• 8 Channel PrismSound ADA-8XR (96kHz/24bit)
• Pitch detection
• Trim and render derivatives for National Jukebox 8
• Harvest and store AES-57 metadata
Figure 1
Figure 1. This is the playback system built for this work. By mounting 4 tone arms, 4 different styli can
be reproducing at the same time. The engineer can compare, in real time, the sound of each stylus and
rapidly switch between them. Compared to the traditional method of manually changing the stylus, then
8 All the work for the Library of Congress is destined for this site:
http : / /www.loc.go v/j ukebox /
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
4
replacing with another, then changing back, etc., there is no need to remember what one stylus sounded
like. With the push a button the operator can compare, not 2 but 4, different styli rapidly.
The Data
The core sources 9 for this study have several attributes that make the data
especially relevant:
• The sample sets are large: more than 9,000 disc sides.
• The data include international sources.
• A period of approximately 20 years of early recording practice is covered.
• All work was performed to the same specifications.
• All work was performed with the same equipment.
• All work was performed by the same engineer.
These are the four data sets used in this study:
• Victor Talking Machine Company
The Victor discs are the stuff collectors, engineers and historians dream of. They are
in pristine condition, beautifully preserved at the Eldridge Johnson Museum. Most of
the rare and international titles in this study are in this collection.
• Edison Diamond Discs
These are factory originals, also in pristine condition [‘reference discs' they might be
called], under the care of Gerald Fabris at the Edison National Historic Park. This
data set includes multiple takes of the same selection.
• OKeh Records
Unlike the Victor and Edison discs, the OKeh disc collection was assembled from
discs that had circulated in the wild. They had been bought in stores, played,
enjoyed, and handled under a random set of circumstances. They are far more
representative of the samples available to most studies. They are holdings of the
Library of Congress, at the Packard Campus of the National Audio-Visual
Conservation Center, in Culpeper, VA.
• Mick Moloney Collection of Irish-American Music and Popular Culture
9 The data for this paper was collected from two contracts performed for the Library of
Congress and another contract for New York University. It is important to note that none of
these contracts required the acquisition of data for this study. Flaws in the methods of data
gathering and analysis are evident throughout. However, the findings are extremely clear,
more than overcoming these flaws. Efforts are taken throughout the paper to point out
where a more disciplined approach could have been followed, and how better method may
or may not have produced different results. The raw data are available at
www.georgeblood.com.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
5
This New York University Collection is included to represent the general population,
as a control and for comparison 10 . Insofar as the focus of the collection is Irish
music, there is no preference for a given label. Like the OKeh Records samples, these
are discs that had circulated in the wild.
The highly uniform digitization process allows us to investigate variation within
each collection and between the four sets, and to explore the sources of those
variations.
The findings from each data set are first presented separately.
PART 1, STYLUS SIZE
Victor Talking Machine Company
The Victor recordings data represents 3,500 sides. When the Victor Talking Machine
Company operated in Camden, NJ, they sent copies of each title to the Free Library
of Philadelphia. Like all libraries, only about 2% of their holdings circulate, and the
sound recordings followed this pattern. When the Library chose to de-accession
their analog sound recordings, the Victor discs moved to the Eldridge Johnson
Museum, run by the Delaware Department of Parks and Recreation. This is an
enormous collection of rarely played discs.
Most of the recordings at the Eldridge Johnson Museum are widely distributed. The
selection of discs to digitize was led by Samuel Brylawski, Editor/Project Manager of
the Encyclopedic Discography of Victor Recordings /American Discography Project
at the University of California at Santa Barbara, someone who we can reasonably
assume to know a thing or two about the Victor catalog. This project selected very
rare recordings of Eastern Europeana, South American, and ethnic recordings.
Therefore, we have an international distribution over a range of two decades,
making our sample set both large, and widely representative.
10 It is important to point out the data from the Moloney Collection used in this study
includes both acoustic and electrical recordings. Therefore, it is far from perfect for
comparison with the Victor, Edison and OKeh data.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
6
Figure 2 illustrates that within our sample set, some time periods are better
represented than others.
This sample bias is the result of selections focused on rare ethnic titles. While this
gives the sample a highly desirable international distribution, the results are less
evenly representative of the time period. A more carefully designed study might
have added samples where the distribution has a low number of titles. However, the
data will show this would not have had a significant impact on the results.
During this period there is great variation of performance practice by region. This is
very important for the investigation of speed and pitch, which will be discussed in
the second half of this paper. The sample set has the advantage that the recordings
were made by teams from the same label.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
7
Stylus Size Distribution
zuuu
1 r r\r\
IbUU
1UUU
rnfi
bUU
n
U
1.5 1.8 2.0 2.3 2.5 2.8 3.3 3.8 4.0
Figure 3 11
Data from the Victor discs show this remarkably clear finding. One stylus size,
2.8mil, was chosen for more than half the Victor sides digitized. The data were next
examined for an explanation of the secondary strengths at 2.0, 2.3 and 2.5mil.
The data were analyzed to determine whether the distribution of sizes matched the
distribution over time. Higher numbers of sides are expected in the subsets at 1916-
1917 and later because there are more data points to begin with (cf. Figure 2).
11 Please pay attention to the scaling on the left side of these charts. Some times it is scaled
for maximum resolution, and other times scaled to a fixed unit to facilitate comparison
between charts.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
8
2.3
Figure 4b
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
9
2.8
300
250
200
150
100
50
ll
, , -I
□
■v
00
o
o
VO
o
00
o o
OV OV
o
t-H
CN
t-H
^ LO
t—H rH
Ov Ov
T— I CN
CN CN
Ov Ov
CO ^ m
CN CN CN
O' o o
Figure 4d
And indeed that is what we see in the data. The peaks in the stylus size subsets
mirror the peaks in the aggregate data. Further, no trend appears in the data. The
data do not show size A being used in the early 19-teens, then a few years later,
toward the 1920’s, a different size is dominant, and so on. The distribution of 2.0,
2.3 and 2.5 is fairly even, with significantly more 2.8 across all time periods.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
10
The data were next analyzed by the recording location. Just as with the distribution
of the stylus size by catalog number, we find the same pattern: all locations
dominated by 2.8 mil, followed by smaller sizes, as shown in Figure 5a-5d.
Camden
Figure 5a
New York
Figure 5b
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
11
Havana
70
60
50
40
30
20
10
0
1.5
.i
3.8 4.0
Figure 5c
In conclusion, when working with acoustic Victor discs, a 95% certainty for proper
stylus size is obtained by reviewing just four stylus sizes. If this sample were
random, that is if it followed the normal distribution under a bell curve, the data
would be symmetrical, with an equal distribution of samples to left and right of the
peak. However, there are nearly lOOx more samples of smaller stylus sizes to the
left of 2.8 than larger stylus sizes to the right of 2.8, indicating something is causing
the non-random distribution.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
12
Might the intended cutting stylus size have been 2.8mil? As the cutter was used, was
it worn down, to a smaller and smaller size, until it was replaced?
Edison Diamond Discs
The Edison recordings are represented by 2,500 sides. Edison’s company retained
samples of most discs it recorded. The collection includes released discs, multiple
takes from which the released version or versions were selected, as well as
unreleased material. Similar in scope and condition to the Victor discs, this is an
enormous collection of rarely played discs.
Edison Stylus Size
lOUU
1600 -
1400 -
1200 -
i r\r\r\
100U
onn
oUU
rr\r\
6UU
400 -
onn
zuu
n
1
1 . _
U
(
iii i i i i
).7 1.5 1.7 1.9
2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9
Stylus Size
Figure 6
The results of analyzing stylus selection for Edison Diamond Discs (Figure 6) are
likewise very clear. They are also very puzzling. Edison, as an engineer, was
meticulous in the specifications for his inventions, their manufacture and use 12 .
Engineers, collectors and historians have long struggled with a great disparity: we
generally agree that Diamond Discs "sounding better”,(whatever that means) when
played with a very small stylus; and yet, the documentation left by Edison indicates
a larger, 3.5 or 3.3 mil, stylus size should be used for playback.
12 Cf. footnote 3.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
13
Disc grooves are three-dimensional: top to bottom, left to right, and along their
length. In practice, either the lateral (in hill and dale) or vertical (in lateral cut)
component is absent 13 . Discussions of stylus shape focus on the cross-sectional view
of the "stylus in the groove”; that is, how the stylus fits and tracks side-to-side or up-
and-down. Less frequent are discussions that consider the relationship between the
longitudinal shape and size of the cutter and that of the playback stylus. The cutter,
by definition, will have a sharp point: it may be 3.5 mil across but it is shaped
(ground or cut) into a sharp point to cut into the recording medium (at first wax,
then later lacquer). A playback stylus that had a sharp edge, and short longitudinal
profile, would naturally tend to cut the disc while being played. Shellac is much
harder than wax, but the problem remains. Further, remember how these media
were played: a stylus wiggled in the groove, then transferred that wiggle to a
diaphragm that transduced the motion into varying air pressure (a.k.a. sound)
which was amplified by a horn - the larger the horn, the greater the amplification.
The stylus both reproduced the sound in the groove, and moved the horn apparatus
across the disc 14 . To keep the stylus in the groove firmly enough to move the horn, a
great deal of pressure was needed. The nominal weight of the diaphragm and the
suspended horn is 4-5 ounces. LP turntables track at below three grams! As the
surface area of the playback stylus is reduced the weight of 4-5 ounces is spread
over a smaller and smaller area. This means more weight is applied to the groove,
leading to increased wear. The sharp cutting stylus incises fine detail along the
groove. The fatter playback stylus, needed to carry the weight of the diaphragm and
horn, cannot capture the information left behind by the cutting stylus. It spans
multiple hills and is unable to reach into the dales 15 .
13 Most discs are cut by modulating the groove side-to-side ("lateral cut”). Edison media are
cut up and down ("vertical cut”), a.k.a. "hill and dale". Vertical cut discs have more
consistent groove geometry. However loud sounds tend to eject the stylus from the groove.
Ultimately lateral cut was used by most manufacturers, and became the standard for LPs.
14 Edison Diamond Disc players include a feed screw to move the stylus and horn across the
disc. This requires the pitch (number of turns per unit distance) of the disc and feed screw
to match. Otherwise the feed screw would cause the disc to skip forward or backward.
15 For a discussion on tracing distortion see Di Toro, J. SMPTE 29, 493, 1937
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
14
Figure 7
Higher frequencies have smaller wavelengths. Smaller wavelengths are smaller groove
sizes (lower waves in the image). Large styli distribute the weight, but don’t fit into the
groove (right). The center stylus fits Edison’s specification. The left stylus sounds
better.
Might it be that Edison’s working assumptions regarding the relationship between
the cutting stylus size and shape 16 and the playback stylus size and shape were
incorrect 17 ? Instead might it be that Edison’s cutting stylus, especially in the hill and
dale topology, resulted in a groove profile best reproduced by a very different
stylus: a size and shape not easily obtained at that time?
16 Edison’s grooved media used a geometry often called "door knob” shape, from the
popular shape of Victorian era hardware. Other manufacturers used conical shape,
triangular in profile, or a triangular shape with the point removed.
17 The assumption is the playback stylus should have the same profile as the cutting stylus.
This is impossible to achieve. The shape might be very, very close, but due to irregularities
in the recording medium, and the replication process, as well as wear of both the cutting
and playback stylus, an exact match cannot be achieved.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
15
OKeh Records
The work on Victor and Edison discs was performed under a contract specifically
intended to generate content for the Library of Congress’ National Jukebox 18 . A later
contract was awarded for digitization of general audiovisual holdings of the Library.
The Moving Image Broadcast and Recorded Sound Division (MBRS) selected a large
span of OKeh 19 Records for preservation digitization under this contract.
The samples are acoustic recordings, the earliest of which are sometimes vertically
cut (like Edisons), spanning approximately a decade. Unlike the Victor and Edison
discs in this study, these discs are heavily worn from general circulation and years
of enjoyment by their owners. Their condition varies widely - very good to poor,
and, rarely, very fine like the Victor and Edison discs. For our study these discs are a
contrasting label and contrasting condition. They are from the same time period,
and have been digitized using the exact methodology as the Victors and the Edisons
(e.g. using four tone arms, etc).
OKeh Stylus Size
-LZ.UU
i nnn
1UUU
onn
oUU
rr\(\ -
6UU
/inn -
4UU
onn
zUU
r\
U
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
0.7 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9
Figure 8
Whereas the Victors showed a clear preference for 2.8mil and Edisons for 0.7mil,
OKeh acoustic discs have an equal preference for 2.3mil. In contrast to where the
18 www.loc.gov/jukebox
19 Fun fact: At the beginning of the 20 th century "OK” was a popular, new expression. While
the spelling evolved to "okay" or simply "OK”, at this time it was spelled "okeh”. The label’s
name is a play on this new expression and the founder’s initials Otto K.E. Heinemann.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
16
Victors show the prominent 2.8 followed by the secondary peaks to the left of
smaller sizes, the OKeh data show an even distribution to the left and right of the
peak, of larger and smaller stylus sizes.
Might the engineers at OKeh have changed cutting styli more frequently than was
the practice at Victor? Despite the clearly audible signs of wear and aging, the OKeh
data on stylus size are among the clearest of all the data sets in this study. Might the
long-held assumption in our trade, that stylus selection includes compensating for
groove wear, be wrong? Could it be that usage wear does not fundamentally warrant
a stylus choice different than would be made on a pristine disc?
Mick Moloney Collection of Irish-American Music and Popular Culture, New
York University
The NYU/Moloney Collection focuses on genre rather than label. The playback stylus
size selection for the discs shown in Figure 9 is included to represent the
distribution of stylus sizes in the general population of discs available throughout
the 78rpm era.
Misc. Labels
Figure 9
This data affirms that the single engineer who digitized all four sets (Victor, Edison,
OKeh and Moloney discs) was thorough and considered multiple stylus sizes during
his selection process. The chart also shows a distribution clearly different from
those from the label-specific data, where there is a clear preference for one size over
all others. Here we see, in a random selection of labels, no clear choice of stylus size
is evident.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
17
All Discs
Z.UUU
loUU
loUU
i/i nn
14UU
i onn
1ZUU
i nnn
■ Victor
1UUU
onn
■ OKeh
oUU
■ Edison
6UU
/inn
4UU
onn
ZUU
r\
■
. i . .
U
C
1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1 1
1.7 1.5 1.7 1.9 2.1 2.3 2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9
Figure 10
The chart in Figure 10 combines all the data from Victor, Edison and OKeh. Both
show many strong peaks and no clear preference, supporting the assertion the
NYU/Moloney collection is representative of the general population, and the label
specific data is clearly different from the general population. The peaks do not
match because the underlying labels are different. The goal is not to say the two
charts match. It is to show that when multiple labels are present in the data there
are multiple peaks. When a single label is examined in large quantities, a clear
preference emerges.
One final data set for stylus selection...
Marcos Sueiro Bal Double Blind Study on Stylus Selection
At the 2015 International Association of Sound and Audiovisual Archives (IASA)
Conference in Paris Marcos Sueiro Bal, Senior Archivist at WNYC in New York and
co-chair of the Association for Recorded Sound Collections (ARSC) Technical
Committee, presented preliminary results of his study on stylus selection. Sueiro Bal
made short digital files of disc transfers. Each disc (presented below in its own
chart. Figures 11a- lie] is a different genre. Each disc was played multiple times,
each time with a different stylus size and shape (A-E or A-F in the following
examples). There were 5 or 6 stylus sizes for each of 5 different musical genres in
preparation for a double blind test, he asked a colleague to rename the samples to
obscure information about the styli used for playback of the samples. Sueiro Bal
asked respected audio transfer engineers to review the files at their leisure using
their preferred playback environment, and to report their preferences. Two
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
18
respondents 20 took the test twice. The responses provided 49 data points for each
musical selection.
The assumption has been that an experienced engineer with training on what to
listen for is needed to choose the proper stylus size; yet, only one test shows a
consensus preference.
The results of the tests are presented in the Figures lla-llf.
Double Blind Stylus Selection Test 1
C
a
o
u
c
o
u
QJ
'o
C/3
Stylus
Figure 11a
20 Marcos Sueiro Bal himself and George Blood, the author of this paper.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
19
Double Blind Stylus Selection Test 2
zu
c 15
3
o
u
C y\ r\
o 10
_u
<u
<u
i/) r
5
0
i i i i ii
A B C D E F
Stylus
Figure lib
Double Blind Stylus Selection Test 3
Stylus
Figure 11c
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
20
Double Blind Stylus Selection Test 4
Stylus
Figure lid
Double Blind Stylus Selection Test 5
zu
£ 15
3
O
u
o 10
V
_U
<U
cu
00 5
n
u
1 1 1 1 1
A B C D E
Stylus
i
F
Figure lie
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
21
Double Blind Stylus Selection Data Stacked
Figure Ilf
The last chart. Figure Ilf, includes all the data points for all five tests together.
There is no distinct preference for any given ordinal choice; that is, there is not a
bias toward choosing the first item heard, for instance. From this distribution it
appears the engineers taking the test took their time to carefully consider all the
alternatives.
The wide range of results for each sample calls into doubt whether we can assert
that the choices of stylus size, even by professionals, results in the "best”, "proper"
or "correct” stylus. This doubt is reinforced by both respondents who took the test
twice and gave different preferences in most instances.
Summary Conclusions on Stylus Size Selection
• The proper stylus for a given label in a given era can be known with high certainty.
- Victor: 2.8mil is the best stylus size. In practice 2.0, 2.3 and 2.5mil sizes are
needed
due to apparent wear in the cutting stylus on some recordings.
- Edison: 0.7mil is the best stylus size.
- OKeh: 2.0 is the best stylus size.
- NYU/Moloney: aggregate data demonstrates a wide range of practice during
the period before 1923.
• Wear may not affect stylus choice after all.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
22
• Stylus selection by experienced and respected engineers is highly variable and not
repeatable according to the generally accepted standards of the scientific method.
PART 2, SPEED DETERMINATION
While pitch and speed are inextricably linked, they will vary independently of each
other during recording. Just as two disc cutters might operate at a different speed,
two artists performing to the same lathe might tune to a different pitch. During
playback however, without additional information, it is not possible to separate the
two effects. The result might be due to the cutting speed, or the tuning pitch, or a
combination of both. Pitch A might be 435Hz or 440Hz. Rotational speed might be
71.29rpm or 78.26rpm. However, if A=440Hz=78.26rpm, then A=435Hz t
71.29rpm.
As discussed in the introduction, speed determination of recordings from the early
20 th century is challenging due to widely varying practices for reference pitch,
difficulty controlling mechanical speed of the recording lathe and playback
turntable, lack of documentation, and other technical and cultural factors 21 .
• Parameters are moving targets.
- A=?
- How fast does a turntable spin?
- How do you control that speed?
- Regional differences in pitch and/or recording practice
- Ambient temperature changes pitch of the instrument between takes
- A work is otherwise too long to fit on one side
- Simply playing out of tune (or different temperament)
- Ego/Ability (want to appear to sing higher or play more "brilliantly”)
- Who cares?
- Operator error (during recording or playback)
As demonstrated for stylus size in the example of the NYU Moloney Collection, our
control group, it is possible the apparent randomness could be the result of many
different variables laid on top of each other. That is, if the data is parsed into its
component subsets, discrete causes might emerge.
21 Other factors may introduce variation or noise into the data. These include digitization
engineer choicing up or down to the nearest semitone, discs whose speed varies across the
duration of the disc, wow and flutter, non-equal tempered tunings, etc. Each of these may
impose additional considerations upon a given side. However, their frequency of
occurrence is low, and therefore only noise in the data, not influencing the conclusions. Cf.
the discussion and charts that follow.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
23
Or they may not. Recall that the distribution of stylus sizes for the Victor discs did
not vary over time or by location.
In the traditional method for pitch detection the engineer matches an external pitch,
whether it be from a tuning fork, pitch pipe or other source, such as a small
keyboard 22 .
Pitch detection for the work in this study used a polychromatic tuner 23 . This
measures all 12 pitches, in real time, at the same time. Pitch is set very rapidly, very
precisely and with high repeatability. This is a significant advance over traditional
methods.
For this project, the clients, the Library of Congress and New York University,
specified that the speed of the discs be varied such that A=440. Noting that
discussion of the merits of that choice are beyond the scope of this paper, the author
wishes only to point out that the choice is highly defensible and, for this study, the
key to producing meaningful data on speed and pitch selection during the acoustic
era. Because, if playback is to be subject to "what I like best”, then the data will vary
just as it did in Marcos Sueiro Bal’s study on stylus selection. There would be no
anchor around which to compare the results.
With the target fixed, investigation of variance from that target is possible. The same
would be true if the target were A=435Hz or rpm=78.26 or 77.92. The locus of the
data changes, but the variation and distribution around the reference does not.
The charts are plotted with percentage deviation from A=440Hz on the vertical axis
(+/- 9.9%), and the number of samples at each 0.1% increment along the horizontal
axis.
22 A few practitioners have perfect pitch. Some choose to leave the discs at 78.2 6rpm and
avoid this topic entirely. Others choose what they like best based on their musical
judgment, and may take into consideration a pitch detection method.
23 PitchLab is available for iPhone iOS and Android smart phones. Rare 12-pitch
Stroboconn and Petersen tuners, no longer manufactured, are also suitable for this work.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
24
Victor Talking Machine Company
Speed Variation Distribution
Figure 12
This result on speed variation in the Victor discs is the exact opposite of the
distribution found in the stylus sizes. Where the stylus data show cleared peaks,
these data are random. They follow normal distribution - also known as a bell curve
- as predicted by the central limit theorem.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
25
Since many of the recordings in this data set were made in remote locations under
difficult circumstances, we can assume more chances for variation in the Victor
recordings. However, as shown below, this distribution is found in all four data sets.
The historical record includes information on pitch over time and by location. The
literature is full of information purporting to establish rotational speed as well. If
either of these were true, we anticipate peaks in the data. Anticipated peaks include
historical pitches, such as frequencies A=415, 430, 435, as well as 440, and proposed
target rpm speeds such as 71.29 and 77.92.
However, when the graph is annotated with these values where we anticipate peaks,
we instead find areas of low sample count. Where there are peaks in the data they
do not correspond to the speed and pitch values proposed by other authors and the
historical record on pitch.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
26
Buenos Aires
Figure 14a
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
27
Camden
Figure 14d
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
28
New York
Mexico City
Camden Buenos Aires
The Victor data were also analyzed by recording location.
The arrows in Figure 14e highlight where the charts appear to indicate some
dominance.
Here again we find the data peaks do not correspond to any anticipated pitch or
speed values, with the exception of one peak in the Mexico City (Figure 14b) data for
A=435, or 17 out of the 3,500 sample points. This is not statistically significant.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
29
Varispeed by Year
1U.UU
♦ «*»♦"*»* ♦♦ m
o.UU
r r\n
♦ ♦ ♦ ♦ ♦
444 ♦♦ ♦♦♦♦♦
444 44 ♦♦♦♦♦♦ ♦
O.UU
a nn
♦ 4m 444 44 ' 44444m
^ ♦ ♦ 4444^4 ♦ >w ^ it g ) ^
4.UU
n nn
♦ ♦ ♦ ♦♦♦♦ AjBgBSSSft ♦♦
Z.UU
n nn
♦
i ♦ ♦♦ 44 « ♦♦♦ #♦
| ft ♦ ♦ »♦ ♦ ♦♦ ♦ ♦♦ HH #♦ ♦
t J 4 4m 44 ♦ #4# #
U.UU
1/0/
i nn
L900 9/26/1902 6/^/1905 ♦*Jl8/19®b 12/yhg^ y^913^ 1/4^19^ l*/25/#2#g>8/n/1924 5/18/1927
-Z.UU
a nn
♦
I ♦♦ ♦ ♦ * ##♦ ♦ 4444444 #»♦ ♦
1 t !♦ ¥«► ♦ h «»► ♦ «*«*
f *f ♦!♦«►♦ ♦♦ ♦ ♦♦
-4.UU
r nn
$ 4 ♦ ♦ ♦ ♦ •
♦ ♦ ♦ #
-O.UU
o nn
♦ ♦ *
-O.UU
V
4
-1U.UU
Figure 15
This scatter plot (Figure 15) distributes the data by catalog number. It clearly
demonstrates a very wide dispersion in each of the subsets of the catalog periods
represented in the data.
The data does show a general slowing of the record speed over approximately 20
years, in that the amount of speed correction necessary to bring the discs to A=440
increases. Slowing the discs allows longer works to be recorded.
An alternative explanation for this trend is that the reference pitch falls over this
period. Since the historical record on pitch is an upward trend, this alternative
explanation seems unlikely.
Edison Diamond Discs
The Library of Congress contract that specified the Victors be pitched to A=440Hz
also specified the Edison Diamond Discs be speed adjusted to 80rpm. This speed
selection is based on strong documentary evidence found in their corporate
archives.
This work on pitch discovery was performed only near the end of the job with the
specific goal of gathering data for this study, rather than as required by the contract.
There are only
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
30
218 data points. An additional 2,200 sides will be digitized, and the data will be
gathered and analyzed in the same manner 24 .
Edison Speed Offset (from 78)
87 I
7.0
5.3
3.6
1.9
0.2
- 1.5
- 3.2
- 4.9
- 6.6
- 8.3
- 10.0
0 10 20 30 40 50 60 70 80
Figure 1 6
These early data show some interesting peaks that correspond to anecdotal
information from other engineers that they often find Diamond Discs to fall not at
80 rpm, but at 78.26 or 79.04rpm.
This data is highly inconclusive.
OKeh Records
Speed variation (Figure 16) in the OKeh discs is also a random distribution.
24 The work on additional Edison Diamond Discs will take place after the publishing
deadline.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
31
8.8
7.1
5.4
I 3 - 7
L
OKe
;h Discs
«
do 2.0
[>
a °-3
£ - 1.4
-3 1
vO
° - 4.8
- 6.5
- 8.2
- 9.9
1
50
0 5
0 1 (
)0 11
50 2 (
)0 21
Figure 1 7
Mick Moloney Collection of Irish-American Music and Popular Culture, New
York University
The data drawn from the random labels in the NYU/Moloney Collection spans a
larger time period than the other data sets, extending well into the electrical era.
Might these data indicate a convergence toward 78.26 as the actual target speed of
discs? Afterall, many of the challenges faced in the acoustic era working with purely
mechanical systems, were either solved or improved over time with the regular use
of reliable and accurate electric motors and standard line frequencies of 50Hz and
60Hz.
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
32
Additional Information on Speed Determination
We embrace that 78.26rpm may not equal A=440Hz, but we want it to equal
something. However, might we be imposing our early 21 st century expectations on a
late 19 th century culture? Consider the following three anecdotes.
First: Attendees of the 2011 IASA Conference in Frankfurt were treated to a
performance by the Sixtonics. The concert included a live recording demonstration
using restored early electrical recording equipment. When the resulting lacquer disc
was played back, one of the musicians asked why it played at a pitch higher than
they had just performed? The drag of the cutting stylus had slowed down the lathe
during recording. When the disc was played back there was less resistance and the
turntable played up to speed.
What is the proper way to play back this disc? At the originally performed
pitch, or the pitch that would have been experienced in the home?
Do we assume it took 100 years for someone to notice this phenomenon? Or
is it merely the case that 100 years ago pitch varied so widely it was taken for
granted?
Might the 79.04 speed observed in the small Edison sample correspond to
this demonstration?
Second: In 2011 your author recorded all 14 Bach keyboard concerti on a collection
of restored antique harpsichords. The instruments were built between 1627 and
1707. When the instruments were built, pitch was trending from A=392 to 415.
During restoration each instrument was pitched differently according to the
tradition of when and where it was originally made. To avoid strain on the
instruments, and many broken strings during the recording, all the instruments
were retuned to a compromise pitch of A=400, a pitch with no historical foundation
whatsoever. Other recordings have been made with these instruments in solo or
continuo roles. In those recordings, made in the span of a few years, the instruments
were tuned to their native pitches: 392, 405, 407, 410 and 415. If these had been
among the Victor, OKeh and Moloney discs in this study pitched to A=440, every one
would be mis-represented, and each in a different way.
To anyone who presumes to know exactly the correct pitch to play any given
acoustic era disc, we offer the following YouTube video.
• https://www.youtube.com/watch?v=UnhlOUBsd6g
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
33
Summary Conclusions on Speed Determination
• Analysis of 4 data sets totaling 9,000 sides, of pristine discs, worn discs and a
random selection of labels, demonstrates no clear pattern to definitively determine
the speed or pitch of acoustic era 78rpm recordings.
These findings inform the process of preservation. It is necessary to use the proper
stylus because it is not possible for the listener to alter that choice. Speed, on the
other hand, can be changed when playing a digital file. In practice it may make sense
to choose a fixed reference (speed or pitch) for playback and leave it to the user to
decide; and to provide multiple playbacks with different styli, and again leave it to
users to chose which they prefer.
Acknowledgements:
John Bolig, Victor discographer extraordinaire
Sam Brylawski, University of California Santa Barbara
Mari Carpenter, Travis Kirspel and Keith Minsinger
Eldridge Johnson Museum, Delaware State Division of Historical and Cultural
Affairs
Brian Destremps, Senior Audio Engineer, George Blood, LP
Caitlin Hunter, Patrick Smetanick, Gene DeAnna, Library of Congress
Morgan Oscar Morel, IT Systems Admin, George Blood, LP
Kimberly Peach, Lead Archivist, The Winthrop Group
Kimberly Tarr, Head, Media Preservation Unit, New York University
PrePrint from forthcoming publication from Indiana University Press of the Proceedings of the Joint
34
